A ferroelectric-ionic-trapping transistor for low power and secure neuromorphic computing

Area of Science:

• * Materials Science and Engineering
• * Computer Engineering
• * Artificial Intelligence Hardware

Background:

• * The increasing demand for energy-efficient and secure computing hardware is driven by the convergence of AI and data analytics.
• * Neuromorphic synaptic devices offer power efficiency through brain-like parallelism but lack inherent data security against malicious read-out.

Purpose of the Study:

• * To introduce a novel device, the ferroelectric-ionic-trapping field-effect transistor (FITFET), capable of both synaptic operations and secure data storage.
• * To develop a hardware-native mechanism for protecting sensitive information in neuromorphic computing systems.

Main Methods:

• * Integration of ferroelectric polarization, oxygen-vacancy migration, and charge trapping within a single transistor structure.
• * Development of two programmable operating regimes: a plain mode for synaptic function and a secure mode for data concealment.
• * Utilizing multiscale analyses and system-level simulations to validate device performance and security features.

Main Results:

• * The FITFET successfully demonstrated dual functionality, enabling fast, low-power synaptic weight modulation and secure data concealment.
• * A reversible, voltage-controlled transition between plain and secure modes was achieved, collapsing the memory window to suppress read attacks.
• * System-level simulations indicated a reduction in model inversion attacks with minimal impact on AI model accuracy.

Conclusions:

• * The FITFET presents a promising solution for energy-efficient and secure neuromorphic computing hardware.
• * The integrated approach offers a hardware-native data protection mechanism, crucial for safeguarding sensitive information in AI systems.